This paper proposes an automatic Chinese text categorization method for solving the emergency event report classification problem. Since bidirectional encoder representations from transformers (BERT) has achieved great success in natural language processing domain, it is employed to derive emergency text features in this study. To overcome the data imbalance problem in the distribution of emergency event categories, a novel loss function is proposed to improve the performance of the BERT-based model. Meanwhile, to avoid the impact of the extreme learning rate, the Adabound optimization algorithm that achieves a gradual smooth transition from Adam to SGD is employed to learn parameters of the model. To verify the feasibility and effectiveness of the proposed method, a Chinese emergency text dataset collected from the Internet is employed. Compared with benchmarking methods, the proposed method has achieved the best performance in terms of accuracy, weighted-precision, weighted-recall, and weighted-F1 values. Therefore, it is promising to employ the proposed method for real applications in smart emergency management systems.